Disclosure of Invention
The invention aims to provide an optical fiber nonlinear equalization method and system based on CNN-biRNN.
In order to solve the technical problems, the invention provides an optical fiber nonlinear equalization method, which comprises the following steps:
collecting data of an optical fiber transmitting end and a receiving end, and processing the data;
constructing a corresponding characteristic sequence according to the processed data;
constructing a training data set and a test data set;
constructing a CNN-biRNN model;
training the CNN-biRNN model according to the training data set; and
and acquiring corresponding recovered transmitting end data according to the test data set through the trained CNN-biRNN model.
Further, the method for collecting the data of the optical fiber transmitting end and the receiving end and processing the data comprises the following steps:
collecting data of an optical fiber transmitting end and a receiving end, performing constellation point label mapping processing on the data of the transmitting end, and performing linear equalization processing on the data of the receiving end, namely
According to a constellation diagram of data of a transmitting end, the constellation diagram contains M constellation points with regularly arranged coordinates, the data of the transmitting end are divided into M different data categories according to M standard constellation points, each standard constellation point corresponds to one data category, and tag numbering is carried out from 1 to M to serve as a tag of the data category;
for the data of the receiving end, the flow of the linear equalization processing comprises the following steps: low pass filtering, amplitude normalization, dispersion compensation, clock recovery, downsampling, orthogonalization processing, polarization demultiplexing and polarization mode dispersion compensation, frequency offset estimation and carrier phase estimation.
Further, the method for constructing the corresponding feature sequence according to the processed data comprises the following steps:
for each code element in the data of the receiving end after linear equalization processing, according to the cross phase modulation in the channel and four wave mixing in the channelFrequency tripletsOr->Constructing their corresponding feature sequences, i.e
The received data sequences after linear equalization processing in the X polarization direction and the Y polarization direction are respectively defined asAnd->To->Corresponds to the kth symbol in the data sequence;
for each symbolTriplet according to intra-channel cross-phase modulation and intra-channel four-wave mixing>Or->Constructing a corresponding characteristic sequence;
corresponding to each code elementWherein the input characteristic sequence comprises complex information of m, n and m+n symbols adjacent to each other in front and back in the data sequence in the polarization direction of the input characteristic sequence and complex information of m, n and m+n symbols adjacent to each other in front and back in the data sequence in the opposite polarization direction;
the kth symbol of the data sequence in the X-polarization directionStructured characteristic sequences->Is thatWherein->Is a two-dimensional vector, is real part data Re and imaginary part data Im of corresponding triplets of cross phase modulation in a channel and four wave mixing in the channel respectively,
for the kth symbol of the data sequence in the Y polarization directionStructured feature sequenceWherein->Is a two-dimensional vector, is real part data Re and imaginary part data Im of corresponding triplets of cross phase modulation in a channel and four wave mixing in the channel respectively,
after the value ranges of m and n are determined, screening is carried out:
where L is a superparameter, determining the maximum of the absolute values of m and n.
Further, the method of constructing a training data set and a test data set includes:
constructing training data set according to preset proportionWherein y is i Is a sequence of featuresCorresponding real labels, label value y epsilon {1,2, …, M };
the remaining data is used as a test data set.
Further, the method for constructing the CNN-birNN model comprises the following steps:
the CNN-birNN model comprises: one-dimensional convolution layer, biRNN layer, flattening layer, full connection layer, softmax layer, and output layer.
Further, the method for training the CNN-biRNN model according to the training data set comprises the following steps:
using the training data setTraining the CNN-birNN model, and determining optimal model parameters by minimizing a cross entropy loss function of a real label and adopting a back propagation algorithm and an Adam parameter optimization algorithm to obtain the trained CNN-birNN model.
Further, the method for acquiring the corresponding recovered transmitting end data according to the test data set by the trained CNN-biRNN model comprises the following steps:
and carrying out label prediction on the test data through a trained CNN-birNN model to obtain a data category corresponding to the current code element, and carrying out label demapping through constellation points to obtain corresponding recovered transmitting end data.
In another aspect, the present invention also provides an optical fiber nonlinear equalization system, including:
the acquisition module acquires data of the optical fiber transmitting end and the optical fiber receiving end and processes the data;
the sequence construction module constructs a corresponding characteristic sequence according to the processed data;
the data set module is used for constructing a training data set and a test data set;
the model building module is used for building a CNN-biRNN model;
the training module is used for training the CNN-biRNN model according to the training data set; and
and the equalization module acquires corresponding restored transmitting end data according to the test data set through the trained CNN-biRNN model.
The invention has the beneficial effects that the data of the optical fiber transmitting end and the receiving end are collected and processed; constructing a corresponding characteristic sequence according to the processed data; constructing a training data set and a test data set; constructing a CNN-biRNN model; training the CNN-biRNN model according to the training data set; and acquiring corresponding recovered transmitting end data according to the test data set through the trained CNN-birNN model, thereby improving the judgment accuracy, realizing nonlinear optical fiber equalization and reducing the bit error rate.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and drawings.
In order to make the above objects, features and advantages of the present invention more comprehensible, preferred embodiments accompanied with figures are described in detail below.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
As shown in fig. 1, the present embodiment provides an optical fiber nonlinear equalization method, which includes: collecting data of an optical fiber transmitting end and a receiving end, and processing the data; constructing a corresponding characteristic sequence according to the processed data; constructing a training data set and a test data set; constructing a CNN-biRNN model; training the CNN-biRNN model according to the training data set; and acquiring corresponding recovered transmitting end data according to the test data set through the trained CNN-birNN model, thereby improving the judgment accuracy, realizing nonlinear optical fiber equalization and reducing the bit error rate. And combining perturbation related information, a Convolutional Neural Network (CNN) and a bidirectional cyclic neural network (birNN) to realize optical fiber nonlinear equalization. The characteristic information of the code elements is enriched by utilizing the triplet of cross phase modulation in the code element channel and four wave mixing in the channel, the characteristic information of the code elements is efficiently extracted by utilizing CNN, the data label of the code elements is predicted by utilizing the advantage of the BiRNN on sequence processing, and the corresponding recovered transmitting end data is obtained through constellation point label demapping, so that the judgment accuracy is improved, the nonlinear equalization of the optical fibers is realized, and the bit error rate is reduced.
Specifically, step 101: data of an optical fiber transmitting end and data of a receiving end are collected, constellation point label mapping processing is carried out on the data of the transmitting end, and linear equalization processing is carried out on the data of the receiving end. Step 102: and constructing a corresponding characteristic sequence of each code element in the data of the receiving end after linear equalization processing according to the triplets of cross phase modulation in the channel and four wave mixing in the channel. Step 103: a training dataset is constructed. Step 104: and constructing a CNN-biRNN model. Step 105: and training the CNN-biRNN model through the constructed training data set, and determining model parameters. Step 106: and carrying out label prediction on the test data through a trained CNN-birNN model, and carrying out label demapping on constellation points to obtain corresponding recovered transmitting end data, thereby improving the judgment accuracy and realizing the nonlinear optical fiber equalization.
In this embodiment, the method for collecting data of the transmitting end and the receiving end of the optical fiber and processing the data includes: collecting data of an optical fiber transmitting end and a receiving end, performing constellation point label mapping processing on the data of the transmitting end, performing linear equalization processing on the data of the receiving end, namely, according to a constellation diagram of the data of the transmitting end, wherein the constellation diagram contains M constellation points with regularly arranged coordinates, the data of the transmitting end are divided into M different data categories according to M standard constellation points, each standard constellation point corresponds to one data category, and tag numbering is performed from 1 to M to serve as a tag of the data category; for the data of the receiving end, the flow of the linear equalization processing comprises the following steps: low pass filtering, amplitude normalization, dispersion compensation, clock recovery, downsampling, orthogonalization processing, polarization demultiplexing and polarization mode dispersion compensation, frequency offset estimation and carrier phase estimation.
In this embodiment, the method for constructing a corresponding feature sequence according to the processed data includes: for each code element in the data of the receiving end after linear equalization processing, according to the triplets of cross phase modulation in the channel and four wave mixing in the channelOr->Constructing their corresponding feature sequences, i.e
The received data sequences after linear equalization processing in the X polarization direction and the Y polarization direction are respectively defined asAnd->To->Corresponds to the kth symbol in the data sequence;
for each symbolTriplet according to intra-channel cross-phase modulation and intra-channel four-wave mixing>Or->Constructing a corresponding characteristic sequence;
corresponding to each code elementWherein the input characteristic sequence comprises complex information of m, n and m+n symbols adjacent to each other in front and back in the data sequence in the polarization direction of the input characteristic sequence and complex information of m, n and m+n symbols adjacent to each other in front and back in the data sequence in the opposite polarization direction;
the kth symbol of the data sequence in the X-polarization directionStructured characteristic sequences->Is thatWherein->Is a two-dimensional vector, is real part data Re and imaginary part data Im of corresponding triplets of cross phase modulation in a channel and four wave mixing in the channel respectively,
for the kth symbol of the data sequence in the Y polarization directionStructured characteristic sequences->Is->Wherein->Is a two-dimensional vector, is real part data Re and imaginary part data Im of corresponding triplets of cross phase modulation in a channel and four wave mixing in the channel respectively,
after the value ranges of m and n are determined, screening is carried out:
where L is a superparameter, determining the maximum of the absolute values of m and n.
In this embodiment, the method for constructing a training data set and a test data set includes: constructing training data set according to preset proportionWherein y is i For->Corresponding real labels, label value y epsilon {1,2, …, M }; the remaining data is used as a test data set.
In this embodiment, the method for constructing the CNN-biRNN model includes: the CNN-birNN model comprises: one-dimensional convolution layer, biRNN layer, flattening layer, full connection layer, softmax layer, and output layer.
In this embodiment, the method for training the CNN-biRNN model according to the training data set includes: using the training data setTraining the CNN-birNN model, and determining optimal model parameters by minimizing a cross entropy loss function of a real label and adopting a back propagation algorithm and an Adam parameter optimization algorithm to obtain the trained CNN-birNN model.
In this embodiment, the method for acquiring corresponding recovered transmitting end data according to the test data set by using the trained CNN-biRNN model includes: and carrying out label prediction on the test data through a trained CNN-birNN model to obtain a data category corresponding to the current code element, and carrying out label demapping through constellation points to obtain corresponding recovered transmitting end data.
In this embodiment, the processes of generating, transmitting, coherent receiving, linear equalization and nonlinear equalization are performed on a 64-QAM signal in a polarization multiplexing 64-QAM coherent optical communication system, so as to verify the effect of combining perturbation related information, convolutional Neural Network (CNN) and bidirectional cyclic neural network (biRNN) to achieve optical fiber nonlinear equalization.
Step 101: the receiving and transmitting of the 64-QAM sequence and the digital signal processing are completed by using MATLAB software and any waveform generator, I/Q modulator, laser, electric amplifier, polarization multiplexing module, erbium-doped fiber amplifier, adjustable optical attenuator, standard single-mode fiber, coherent receiver, real-time oscilloscope and other instruments.
Carrying out constellation point label mapping on 64-QAM signal data of a transmitting end, wherein a constellation diagram of a 64-QAM modulation format signal has 64 standard constellation points, the 64-QAM signal is divided into 64 different categories according to the 64 standard constellation points, each standard constellation point corresponds to one category, and numbering is carried out from 1 to 64 to serve as a label of a data category; after passing through the optical fiber transmission system, the signal is damaged by a series of linearity and nonlinearity in the transmission process; and carrying out linear equalization processing on the 64-QAM signal data received at a receiving end, wherein the linear equalization processing comprises low-pass filtering, I/Q imbalance compensation, dispersion compensation, clock recovery, polarization demultiplexing, polarization mode dispersion compensation, frequency offset estimation and carrier recovery.
Step 102: for each code element in the data of the receiving end after linear equalization processing, according to the triplets of cross phase modulation in the channel and four wave mixing in the channelOr->And constructing a corresponding characteristic sequence. The kth symbol for the data sequence in the X polarization direction +.>Structured characteristic sequences->Is->Wherein->Is a two-dimensional vector, which is real part (Re) data and imaginary part (Im) data of corresponding triplets of cross phase modulation in a channel and four-wave mixing in the channel respectively,
for the Y polarization directionK-th symbol of data sequence->Structured characteristic sequences->Is thatWherein->Is a two-dimensional vector, which is real part (Re) data and imaginary part (Im) data of corresponding triplets of cross phase modulation in a channel and four-wave mixing in the channel respectively,reasonably selecting the value ranges of m and n, and screening according to the following rules: />Where L is a superparameter, determining the maximum of the absolute values of m and n. The super parameter L has a value of 75, and the m has a value range of [ -75,75]The value range of n is [ -75,75]The number of combinations of values of m and n that satisfy the condition is 1629, and therefore, the length of the characteristic sequence of each symbol is 1629.
Step 103: a training dataset is constructed. Training data is randomly selected according to a certain proportion, and a training data set is constructedWherein y is i For->The corresponding real tag has tag value y e {1,2, …,64}. The remainder served as the test data.
Step 104: and constructing a CNN-biRNN model. The CNN-birNN model comprises: a one-dimensional convolutional layer, a biRNN layer, a flattening layer, a fully-connected layer, a Softmax layer, and an output layer, as shown in fig. 2.
In this embodiment, the length of the feature sequence of each symbol is 1629, each vector in the feature sequence is a two-dimensional vector, batch processing size batch is 128, a matrix with one dimension (128, 1629,2) is input to a one-dimensional convolution layer, padding calculation is performed on the one-dimensional convolution layer and 16 convolution kernels with one dimension (128,3,2), the matrix is converted into a matrix with one dimension (128, 1629, 16), the matrix is input to a biRNN layer for processing, the matrix with one dimension (128, 1629, 64) is input to a flattening layer for processing, the matrix with one dimension (128, 104256) is output to a fully connected layer for processing, the matrix with one dimension (128, 64) is input to a Softmax layer, the output of the Softmax layer is the probability that the current symbol belongs to each data category, and the calculation formula is as follows:
finally, the output layer outputs a matrix with the dimension of (128, 1) according to the maximum probability, wherein 128 tag values are tag values corresponding to the symbol data corresponding to the batch.
Step 105: and training the CNN-biRNN model through the constructed training data set, and determining model parameters. Training the CNN-birNN model by using the training data set constructed in the step 103, and determining optimal model parameters by minimizing a cross entropy loss function of a real label and adopting a back propagation algorithm and an Adam parameter optimization algorithm to obtain the trained CNN-birNN model.
Step 106: and carrying out label prediction on the test data through a trained CNN-birNN model, and carrying out label demapping on constellation points to obtain corresponding recovered transmitting end data, thereby improving the judgment accuracy and realizing the nonlinear optical fiber equalization.
There is also provided in this embodiment an optical fiber nonlinear equalization system, including: the acquisition module acquires data of the optical fiber transmitting end and the optical fiber receiving end and processes the data; the sequence construction module constructs a corresponding characteristic sequence according to the processed data; the data set module is used for constructing a training data set and a test data set; the model building module is used for building a CNN-biRNN model; the training module is used for training the CNN-biRNN model according to the training data set; the equalization module acquires corresponding restored transmitting end data according to the test data set through the trained CNN-biRNN model; the specific functions of each module have been described in detail, and will not be described in detail in this embodiment.
In summary, the invention constructs the training data set by collecting the data of the optical fiber transmitting end and the receiving end and using the characteristic sequence of the ternary combination data of the cross phase modulation in the channel and the four wave mixing in the channel; constructing a nonlinear equalization model based on CNN-biRNN; training the CNN-biRNN model by using a training data set; and (3) carrying out label prediction on the characteristic sequence of each code element in the test set by using a trained CNN-birNN model, and carrying out label demapping on the constellation point to obtain corresponding recovered transmitting end data, thereby improving the judgment accuracy, realizing the nonlinear equalization of the optical fiber and reducing the bit error rate.
In the several embodiments provided in this application, it should be understood that the disclosed apparatus and method may be implemented in other manners as well. The apparatus embodiments described above are merely illustrative, for example, of the flowcharts and block diagrams in the figures that illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
In addition, functional modules in the embodiments of the present invention may be integrated together to form a single part, or each module may exist alone, or two or more modules may be integrated to form a single part.
The functions, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
With the above-described preferred embodiments according to the present invention as an illustration, the above-described descriptions can be used by persons skilled in the relevant art to make various changes and modifications without departing from the scope of the technical idea of the present invention. The technical scope of the present invention is not limited to the description, but must be determined according to the scope of claims.